1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (C) 2011 STRATO. All rights reserved.
4 */
5
6 #include <linux/sched.h>
7 #include <linux/pagemap.h>
8 #include <linux/writeback.h>
9 #include <linux/blkdev.h>
10 #include <linux/slab.h>
11 #include <linux/workqueue.h>
12 #include "ctree.h"
13 #include "volumes.h"
14 #include "disk-io.h"
15 #include "transaction.h"
16 #include "dev-replace.h"
17
18 #undef DEBUG
19
20 /*
21 * This is the implementation for the generic read ahead framework.
22 *
23 * To trigger a readahead, btrfs_reada_add must be called. It will start
24 * a read ahead for the given range [start, end) on tree root. The returned
25 * handle can either be used to wait on the readahead to finish
26 * (btrfs_reada_wait), or to send it to the background (btrfs_reada_detach).
27 *
28 * The read ahead works as follows:
29 * On btrfs_reada_add, the root of the tree is inserted into a radix_tree.
30 * reada_start_machine will then search for extents to prefetch and trigger
31 * some reads. When a read finishes for a node, all contained node/leaf
32 * pointers that lie in the given range will also be enqueued. The reads will
33 * be triggered in sequential order, thus giving a big win over a naive
34 * enumeration. It will also make use of multi-device layouts. Each disk
35 * will have its on read pointer and all disks will by utilized in parallel.
36 * Also will no two disks read both sides of a mirror simultaneously, as this
37 * would waste seeking capacity. Instead both disks will read different parts
38 * of the filesystem.
39 * Any number of readaheads can be started in parallel. The read order will be
40 * determined globally, i.e. 2 parallel readaheads will normally finish faster
41 * than the 2 started one after another.
42 */
43
44 #define MAX_IN_FLIGHT 6
45
46 struct reada_extctl {
47 struct list_head list;
48 struct reada_control *rc;
49 u64 generation;
50 };
51
52 struct reada_extent {
53 u64 logical;
54 struct btrfs_key top;
55 struct list_head extctl;
56 int refcnt;
57 spinlock_t lock;
58 struct reada_zone *zones[BTRFS_MAX_MIRRORS];
59 int nzones;
60 int scheduled;
61 };
62
63 struct reada_zone {
64 u64 start;
65 u64 end;
66 u64 elems;
67 struct list_head list;
68 spinlock_t lock;
69 int locked;
70 struct btrfs_device *device;
71 struct btrfs_device *devs[BTRFS_MAX_MIRRORS]; /* full list, incl
72 * self */
73 int ndevs;
74 struct kref refcnt;
75 };
76
77 struct reada_machine_work {
78 struct btrfs_work work;
79 struct btrfs_fs_info *fs_info;
80 };
81
82 static void reada_extent_put(struct btrfs_fs_info *, struct reada_extent *);
83 static void reada_control_release(struct kref *kref);
84 static void reada_zone_release(struct kref *kref);
85 static void reada_start_machine(struct btrfs_fs_info *fs_info);
86 static void __reada_start_machine(struct btrfs_fs_info *fs_info);
87
88 static int reada_add_block(struct reada_control *rc, u64 logical,
89 struct btrfs_key *top, u64 generation);
90
91 /* recurses */
92 /* in case of err, eb might be NULL */
__readahead_hook(struct btrfs_fs_info * fs_info,struct reada_extent * re,struct extent_buffer * eb,int err)93 static void __readahead_hook(struct btrfs_fs_info *fs_info,
94 struct reada_extent *re, struct extent_buffer *eb,
95 int err)
96 {
97 int nritems;
98 int i;
99 u64 bytenr;
100 u64 generation;
101 struct list_head list;
102
103 spin_lock(&re->lock);
104 /*
105 * just take the full list from the extent. afterwards we
106 * don't need the lock anymore
107 */
108 list_replace_init(&re->extctl, &list);
109 re->scheduled = 0;
110 spin_unlock(&re->lock);
111
112 /*
113 * this is the error case, the extent buffer has not been
114 * read correctly. We won't access anything from it and
115 * just cleanup our data structures. Effectively this will
116 * cut the branch below this node from read ahead.
117 */
118 if (err)
119 goto cleanup;
120
121 /*
122 * FIXME: currently we just set nritems to 0 if this is a leaf,
123 * effectively ignoring the content. In a next step we could
124 * trigger more readahead depending from the content, e.g.
125 * fetch the checksums for the extents in the leaf.
126 */
127 if (!btrfs_header_level(eb))
128 goto cleanup;
129
130 nritems = btrfs_header_nritems(eb);
131 generation = btrfs_header_generation(eb);
132 for (i = 0; i < nritems; i++) {
133 struct reada_extctl *rec;
134 u64 n_gen;
135 struct btrfs_key key;
136 struct btrfs_key next_key;
137
138 btrfs_node_key_to_cpu(eb, &key, i);
139 if (i + 1 < nritems)
140 btrfs_node_key_to_cpu(eb, &next_key, i + 1);
141 else
142 next_key = re->top;
143 bytenr = btrfs_node_blockptr(eb, i);
144 n_gen = btrfs_node_ptr_generation(eb, i);
145
146 list_for_each_entry(rec, &list, list) {
147 struct reada_control *rc = rec->rc;
148
149 /*
150 * if the generation doesn't match, just ignore this
151 * extctl. This will probably cut off a branch from
152 * prefetch. Alternatively one could start a new (sub-)
153 * prefetch for this branch, starting again from root.
154 * FIXME: move the generation check out of this loop
155 */
156 #ifdef DEBUG
157 if (rec->generation != generation) {
158 btrfs_debug(fs_info,
159 "generation mismatch for (%llu,%d,%llu) %llu != %llu",
160 key.objectid, key.type, key.offset,
161 rec->generation, generation);
162 }
163 #endif
164 if (rec->generation == generation &&
165 btrfs_comp_cpu_keys(&key, &rc->key_end) < 0 &&
166 btrfs_comp_cpu_keys(&next_key, &rc->key_start) > 0)
167 reada_add_block(rc, bytenr, &next_key, n_gen);
168 }
169 }
170
171 cleanup:
172 /*
173 * free extctl records
174 */
175 while (!list_empty(&list)) {
176 struct reada_control *rc;
177 struct reada_extctl *rec;
178
179 rec = list_first_entry(&list, struct reada_extctl, list);
180 list_del(&rec->list);
181 rc = rec->rc;
182 kfree(rec);
183
184 kref_get(&rc->refcnt);
185 if (atomic_dec_and_test(&rc->elems)) {
186 kref_put(&rc->refcnt, reada_control_release);
187 wake_up(&rc->wait);
188 }
189 kref_put(&rc->refcnt, reada_control_release);
190
191 reada_extent_put(fs_info, re); /* one ref for each entry */
192 }
193
194 return;
195 }
196
btree_readahead_hook(struct extent_buffer * eb,int err)197 int btree_readahead_hook(struct extent_buffer *eb, int err)
198 {
199 struct btrfs_fs_info *fs_info = eb->fs_info;
200 int ret = 0;
201 struct reada_extent *re;
202
203 /* find extent */
204 spin_lock(&fs_info->reada_lock);
205 re = radix_tree_lookup(&fs_info->reada_tree,
206 eb->start >> PAGE_SHIFT);
207 if (re)
208 re->refcnt++;
209 spin_unlock(&fs_info->reada_lock);
210 if (!re) {
211 ret = -1;
212 goto start_machine;
213 }
214
215 __readahead_hook(fs_info, re, eb, err);
216 reada_extent_put(fs_info, re); /* our ref */
217
218 start_machine:
219 reada_start_machine(fs_info);
220 return ret;
221 }
222
reada_find_zone(struct btrfs_device * dev,u64 logical,struct btrfs_bio * bbio)223 static struct reada_zone *reada_find_zone(struct btrfs_device *dev, u64 logical,
224 struct btrfs_bio *bbio)
225 {
226 struct btrfs_fs_info *fs_info = dev->fs_info;
227 int ret;
228 struct reada_zone *zone;
229 struct btrfs_block_group_cache *cache = NULL;
230 u64 start;
231 u64 end;
232 int i;
233
234 zone = NULL;
235 spin_lock(&fs_info->reada_lock);
236 ret = radix_tree_gang_lookup(&dev->reada_zones, (void **)&zone,
237 logical >> PAGE_SHIFT, 1);
238 if (ret == 1 && logical >= zone->start && logical <= zone->end) {
239 kref_get(&zone->refcnt);
240 spin_unlock(&fs_info->reada_lock);
241 return zone;
242 }
243
244 spin_unlock(&fs_info->reada_lock);
245
246 cache = btrfs_lookup_block_group(fs_info, logical);
247 if (!cache)
248 return NULL;
249
250 start = cache->key.objectid;
251 end = start + cache->key.offset - 1;
252 btrfs_put_block_group(cache);
253
254 zone = kzalloc(sizeof(*zone), GFP_KERNEL);
255 if (!zone)
256 return NULL;
257
258 ret = radix_tree_preload(GFP_KERNEL);
259 if (ret) {
260 kfree(zone);
261 return NULL;
262 }
263
264 zone->start = start;
265 zone->end = end;
266 INIT_LIST_HEAD(&zone->list);
267 spin_lock_init(&zone->lock);
268 zone->locked = 0;
269 kref_init(&zone->refcnt);
270 zone->elems = 0;
271 zone->device = dev; /* our device always sits at index 0 */
272 for (i = 0; i < bbio->num_stripes; ++i) {
273 /* bounds have already been checked */
274 zone->devs[i] = bbio->stripes[i].dev;
275 }
276 zone->ndevs = bbio->num_stripes;
277
278 spin_lock(&fs_info->reada_lock);
279 ret = radix_tree_insert(&dev->reada_zones,
280 (unsigned long)(zone->end >> PAGE_SHIFT),
281 zone);
282
283 if (ret == -EEXIST) {
284 kfree(zone);
285 ret = radix_tree_gang_lookup(&dev->reada_zones, (void **)&zone,
286 logical >> PAGE_SHIFT, 1);
287 if (ret == 1 && logical >= zone->start && logical <= zone->end)
288 kref_get(&zone->refcnt);
289 else
290 zone = NULL;
291 }
292 spin_unlock(&fs_info->reada_lock);
293 radix_tree_preload_end();
294
295 return zone;
296 }
297
reada_find_extent(struct btrfs_fs_info * fs_info,u64 logical,struct btrfs_key * top)298 static struct reada_extent *reada_find_extent(struct btrfs_fs_info *fs_info,
299 u64 logical,
300 struct btrfs_key *top)
301 {
302 int ret;
303 struct reada_extent *re = NULL;
304 struct reada_extent *re_exist = NULL;
305 struct btrfs_bio *bbio = NULL;
306 struct btrfs_device *dev;
307 struct btrfs_device *prev_dev;
308 u64 length;
309 int real_stripes;
310 int nzones = 0;
311 unsigned long index = logical >> PAGE_SHIFT;
312 int dev_replace_is_ongoing;
313 int have_zone = 0;
314
315 spin_lock(&fs_info->reada_lock);
316 re = radix_tree_lookup(&fs_info->reada_tree, index);
317 if (re)
318 re->refcnt++;
319 spin_unlock(&fs_info->reada_lock);
320
321 if (re)
322 return re;
323
324 re = kzalloc(sizeof(*re), GFP_KERNEL);
325 if (!re)
326 return NULL;
327
328 re->logical = logical;
329 re->top = *top;
330 INIT_LIST_HEAD(&re->extctl);
331 spin_lock_init(&re->lock);
332 re->refcnt = 1;
333
334 /*
335 * map block
336 */
337 length = fs_info->nodesize;
338 ret = btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, logical,
339 &length, &bbio, 0);
340 if (ret || !bbio || length < fs_info->nodesize)
341 goto error;
342
343 if (bbio->num_stripes > BTRFS_MAX_MIRRORS) {
344 btrfs_err(fs_info,
345 "readahead: more than %d copies not supported",
346 BTRFS_MAX_MIRRORS);
347 goto error;
348 }
349
350 real_stripes = bbio->num_stripes - bbio->num_tgtdevs;
351 for (nzones = 0; nzones < real_stripes; ++nzones) {
352 struct reada_zone *zone;
353
354 dev = bbio->stripes[nzones].dev;
355
356 /* cannot read ahead on missing device. */
357 if (!dev->bdev)
358 continue;
359
360 zone = reada_find_zone(dev, logical, bbio);
361 if (!zone)
362 continue;
363
364 re->zones[re->nzones++] = zone;
365 spin_lock(&zone->lock);
366 if (!zone->elems)
367 kref_get(&zone->refcnt);
368 ++zone->elems;
369 spin_unlock(&zone->lock);
370 spin_lock(&fs_info->reada_lock);
371 kref_put(&zone->refcnt, reada_zone_release);
372 spin_unlock(&fs_info->reada_lock);
373 }
374 if (re->nzones == 0) {
375 /* not a single zone found, error and out */
376 goto error;
377 }
378
379 ret = radix_tree_preload(GFP_KERNEL);
380 if (ret)
381 goto error;
382
383 /* insert extent in reada_tree + all per-device trees, all or nothing */
384 btrfs_dev_replace_read_lock(&fs_info->dev_replace);
385 spin_lock(&fs_info->reada_lock);
386 ret = radix_tree_insert(&fs_info->reada_tree, index, re);
387 if (ret == -EEXIST) {
388 re_exist = radix_tree_lookup(&fs_info->reada_tree, index);
389 re_exist->refcnt++;
390 spin_unlock(&fs_info->reada_lock);
391 btrfs_dev_replace_read_unlock(&fs_info->dev_replace);
392 radix_tree_preload_end();
393 goto error;
394 }
395 if (ret) {
396 spin_unlock(&fs_info->reada_lock);
397 btrfs_dev_replace_read_unlock(&fs_info->dev_replace);
398 radix_tree_preload_end();
399 goto error;
400 }
401 radix_tree_preload_end();
402 prev_dev = NULL;
403 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(
404 &fs_info->dev_replace);
405 for (nzones = 0; nzones < re->nzones; ++nzones) {
406 dev = re->zones[nzones]->device;
407
408 if (dev == prev_dev) {
409 /*
410 * in case of DUP, just add the first zone. As both
411 * are on the same device, there's nothing to gain
412 * from adding both.
413 * Also, it wouldn't work, as the tree is per device
414 * and adding would fail with EEXIST
415 */
416 continue;
417 }
418 if (!dev->bdev)
419 continue;
420
421 if (dev_replace_is_ongoing &&
422 dev == fs_info->dev_replace.tgtdev) {
423 /*
424 * as this device is selected for reading only as
425 * a last resort, skip it for read ahead.
426 */
427 continue;
428 }
429 prev_dev = dev;
430 ret = radix_tree_insert(&dev->reada_extents, index, re);
431 if (ret) {
432 while (--nzones >= 0) {
433 dev = re->zones[nzones]->device;
434 BUG_ON(dev == NULL);
435 /* ignore whether the entry was inserted */
436 radix_tree_delete(&dev->reada_extents, index);
437 }
438 radix_tree_delete(&fs_info->reada_tree, index);
439 spin_unlock(&fs_info->reada_lock);
440 btrfs_dev_replace_read_unlock(&fs_info->dev_replace);
441 goto error;
442 }
443 have_zone = 1;
444 }
445 if (!have_zone)
446 radix_tree_delete(&fs_info->reada_tree, index);
447 spin_unlock(&fs_info->reada_lock);
448 btrfs_dev_replace_read_unlock(&fs_info->dev_replace);
449
450 if (!have_zone)
451 goto error;
452
453 btrfs_put_bbio(bbio);
454 return re;
455
456 error:
457 for (nzones = 0; nzones < re->nzones; ++nzones) {
458 struct reada_zone *zone;
459
460 zone = re->zones[nzones];
461 kref_get(&zone->refcnt);
462 spin_lock(&zone->lock);
463 --zone->elems;
464 if (zone->elems == 0) {
465 /*
466 * no fs_info->reada_lock needed, as this can't be
467 * the last ref
468 */
469 kref_put(&zone->refcnt, reada_zone_release);
470 }
471 spin_unlock(&zone->lock);
472
473 spin_lock(&fs_info->reada_lock);
474 kref_put(&zone->refcnt, reada_zone_release);
475 spin_unlock(&fs_info->reada_lock);
476 }
477 btrfs_put_bbio(bbio);
478 kfree(re);
479 return re_exist;
480 }
481
reada_extent_put(struct btrfs_fs_info * fs_info,struct reada_extent * re)482 static void reada_extent_put(struct btrfs_fs_info *fs_info,
483 struct reada_extent *re)
484 {
485 int i;
486 unsigned long index = re->logical >> PAGE_SHIFT;
487
488 spin_lock(&fs_info->reada_lock);
489 if (--re->refcnt) {
490 spin_unlock(&fs_info->reada_lock);
491 return;
492 }
493
494 radix_tree_delete(&fs_info->reada_tree, index);
495 for (i = 0; i < re->nzones; ++i) {
496 struct reada_zone *zone = re->zones[i];
497
498 radix_tree_delete(&zone->device->reada_extents, index);
499 }
500
501 spin_unlock(&fs_info->reada_lock);
502
503 for (i = 0; i < re->nzones; ++i) {
504 struct reada_zone *zone = re->zones[i];
505
506 kref_get(&zone->refcnt);
507 spin_lock(&zone->lock);
508 --zone->elems;
509 if (zone->elems == 0) {
510 /* no fs_info->reada_lock needed, as this can't be
511 * the last ref */
512 kref_put(&zone->refcnt, reada_zone_release);
513 }
514 spin_unlock(&zone->lock);
515
516 spin_lock(&fs_info->reada_lock);
517 kref_put(&zone->refcnt, reada_zone_release);
518 spin_unlock(&fs_info->reada_lock);
519 }
520
521 kfree(re);
522 }
523
reada_zone_release(struct kref * kref)524 static void reada_zone_release(struct kref *kref)
525 {
526 struct reada_zone *zone = container_of(kref, struct reada_zone, refcnt);
527
528 radix_tree_delete(&zone->device->reada_zones,
529 zone->end >> PAGE_SHIFT);
530
531 kfree(zone);
532 }
533
reada_control_release(struct kref * kref)534 static void reada_control_release(struct kref *kref)
535 {
536 struct reada_control *rc = container_of(kref, struct reada_control,
537 refcnt);
538
539 kfree(rc);
540 }
541
reada_add_block(struct reada_control * rc,u64 logical,struct btrfs_key * top,u64 generation)542 static int reada_add_block(struct reada_control *rc, u64 logical,
543 struct btrfs_key *top, u64 generation)
544 {
545 struct btrfs_fs_info *fs_info = rc->fs_info;
546 struct reada_extent *re;
547 struct reada_extctl *rec;
548
549 /* takes one ref */
550 re = reada_find_extent(fs_info, logical, top);
551 if (!re)
552 return -1;
553
554 rec = kzalloc(sizeof(*rec), GFP_KERNEL);
555 if (!rec) {
556 reada_extent_put(fs_info, re);
557 return -ENOMEM;
558 }
559
560 rec->rc = rc;
561 rec->generation = generation;
562 atomic_inc(&rc->elems);
563
564 spin_lock(&re->lock);
565 list_add_tail(&rec->list, &re->extctl);
566 spin_unlock(&re->lock);
567
568 /* leave the ref on the extent */
569
570 return 0;
571 }
572
573 /*
574 * called with fs_info->reada_lock held
575 */
reada_peer_zones_set_lock(struct reada_zone * zone,int lock)576 static void reada_peer_zones_set_lock(struct reada_zone *zone, int lock)
577 {
578 int i;
579 unsigned long index = zone->end >> PAGE_SHIFT;
580
581 for (i = 0; i < zone->ndevs; ++i) {
582 struct reada_zone *peer;
583 peer = radix_tree_lookup(&zone->devs[i]->reada_zones, index);
584 if (peer && peer->device != zone->device)
585 peer->locked = lock;
586 }
587 }
588
589 /*
590 * called with fs_info->reada_lock held
591 */
reada_pick_zone(struct btrfs_device * dev)592 static int reada_pick_zone(struct btrfs_device *dev)
593 {
594 struct reada_zone *top_zone = NULL;
595 struct reada_zone *top_locked_zone = NULL;
596 u64 top_elems = 0;
597 u64 top_locked_elems = 0;
598 unsigned long index = 0;
599 int ret;
600
601 if (dev->reada_curr_zone) {
602 reada_peer_zones_set_lock(dev->reada_curr_zone, 0);
603 kref_put(&dev->reada_curr_zone->refcnt, reada_zone_release);
604 dev->reada_curr_zone = NULL;
605 }
606 /* pick the zone with the most elements */
607 while (1) {
608 struct reada_zone *zone;
609
610 ret = radix_tree_gang_lookup(&dev->reada_zones,
611 (void **)&zone, index, 1);
612 if (ret == 0)
613 break;
614 index = (zone->end >> PAGE_SHIFT) + 1;
615 if (zone->locked) {
616 if (zone->elems > top_locked_elems) {
617 top_locked_elems = zone->elems;
618 top_locked_zone = zone;
619 }
620 } else {
621 if (zone->elems > top_elems) {
622 top_elems = zone->elems;
623 top_zone = zone;
624 }
625 }
626 }
627 if (top_zone)
628 dev->reada_curr_zone = top_zone;
629 else if (top_locked_zone)
630 dev->reada_curr_zone = top_locked_zone;
631 else
632 return 0;
633
634 dev->reada_next = dev->reada_curr_zone->start;
635 kref_get(&dev->reada_curr_zone->refcnt);
636 reada_peer_zones_set_lock(dev->reada_curr_zone, 1);
637
638 return 1;
639 }
640
reada_start_machine_dev(struct btrfs_device * dev)641 static int reada_start_machine_dev(struct btrfs_device *dev)
642 {
643 struct btrfs_fs_info *fs_info = dev->fs_info;
644 struct reada_extent *re = NULL;
645 int mirror_num = 0;
646 struct extent_buffer *eb = NULL;
647 u64 logical;
648 int ret;
649 int i;
650
651 spin_lock(&fs_info->reada_lock);
652 if (dev->reada_curr_zone == NULL) {
653 ret = reada_pick_zone(dev);
654 if (!ret) {
655 spin_unlock(&fs_info->reada_lock);
656 return 0;
657 }
658 }
659 /*
660 * FIXME currently we issue the reads one extent at a time. If we have
661 * a contiguous block of extents, we could also coagulate them or use
662 * plugging to speed things up
663 */
664 ret = radix_tree_gang_lookup(&dev->reada_extents, (void **)&re,
665 dev->reada_next >> PAGE_SHIFT, 1);
666 if (ret == 0 || re->logical > dev->reada_curr_zone->end) {
667 ret = reada_pick_zone(dev);
668 if (!ret) {
669 spin_unlock(&fs_info->reada_lock);
670 return 0;
671 }
672 re = NULL;
673 ret = radix_tree_gang_lookup(&dev->reada_extents, (void **)&re,
674 dev->reada_next >> PAGE_SHIFT, 1);
675 }
676 if (ret == 0) {
677 spin_unlock(&fs_info->reada_lock);
678 return 0;
679 }
680 dev->reada_next = re->logical + fs_info->nodesize;
681 re->refcnt++;
682
683 spin_unlock(&fs_info->reada_lock);
684
685 spin_lock(&re->lock);
686 if (re->scheduled || list_empty(&re->extctl)) {
687 spin_unlock(&re->lock);
688 reada_extent_put(fs_info, re);
689 return 0;
690 }
691 re->scheduled = 1;
692 spin_unlock(&re->lock);
693
694 /*
695 * find mirror num
696 */
697 for (i = 0; i < re->nzones; ++i) {
698 if (re->zones[i]->device == dev) {
699 mirror_num = i + 1;
700 break;
701 }
702 }
703 logical = re->logical;
704
705 atomic_inc(&dev->reada_in_flight);
706 ret = reada_tree_block_flagged(fs_info, logical, mirror_num, &eb);
707 if (ret)
708 __readahead_hook(fs_info, re, NULL, ret);
709 else if (eb)
710 __readahead_hook(fs_info, re, eb, ret);
711
712 if (eb)
713 free_extent_buffer(eb);
714
715 atomic_dec(&dev->reada_in_flight);
716 reada_extent_put(fs_info, re);
717
718 return 1;
719
720 }
721
reada_start_machine_worker(struct btrfs_work * work)722 static void reada_start_machine_worker(struct btrfs_work *work)
723 {
724 struct reada_machine_work *rmw;
725 int old_ioprio;
726
727 rmw = container_of(work, struct reada_machine_work, work);
728
729 old_ioprio = IOPRIO_PRIO_VALUE(task_nice_ioclass(current),
730 task_nice_ioprio(current));
731 set_task_ioprio(current, BTRFS_IOPRIO_READA);
732 __reada_start_machine(rmw->fs_info);
733 set_task_ioprio(current, old_ioprio);
734
735 atomic_dec(&rmw->fs_info->reada_works_cnt);
736
737 kfree(rmw);
738 }
739
__reada_start_machine(struct btrfs_fs_info * fs_info)740 static void __reada_start_machine(struct btrfs_fs_info *fs_info)
741 {
742 struct btrfs_device *device;
743 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
744 u64 enqueued;
745 u64 total = 0;
746 int i;
747
748 again:
749 do {
750 enqueued = 0;
751 mutex_lock(&fs_devices->device_list_mutex);
752 list_for_each_entry(device, &fs_devices->devices, dev_list) {
753 if (atomic_read(&device->reada_in_flight) <
754 MAX_IN_FLIGHT)
755 enqueued += reada_start_machine_dev(device);
756 }
757 mutex_unlock(&fs_devices->device_list_mutex);
758 total += enqueued;
759 } while (enqueued && total < 10000);
760 if (fs_devices->seed) {
761 fs_devices = fs_devices->seed;
762 goto again;
763 }
764
765 if (enqueued == 0)
766 return;
767
768 /*
769 * If everything is already in the cache, this is effectively single
770 * threaded. To a) not hold the caller for too long and b) to utilize
771 * more cores, we broke the loop above after 10000 iterations and now
772 * enqueue to workers to finish it. This will distribute the load to
773 * the cores.
774 */
775 for (i = 0; i < 2; ++i) {
776 reada_start_machine(fs_info);
777 if (atomic_read(&fs_info->reada_works_cnt) >
778 BTRFS_MAX_MIRRORS * 2)
779 break;
780 }
781 }
782
reada_start_machine(struct btrfs_fs_info * fs_info)783 static void reada_start_machine(struct btrfs_fs_info *fs_info)
784 {
785 struct reada_machine_work *rmw;
786
787 rmw = kzalloc(sizeof(*rmw), GFP_KERNEL);
788 if (!rmw) {
789 /* FIXME we cannot handle this properly right now */
790 BUG();
791 }
792 btrfs_init_work(&rmw->work, btrfs_readahead_helper,
793 reada_start_machine_worker, NULL, NULL);
794 rmw->fs_info = fs_info;
795
796 btrfs_queue_work(fs_info->readahead_workers, &rmw->work);
797 atomic_inc(&fs_info->reada_works_cnt);
798 }
799
800 #ifdef DEBUG
dump_devs(struct btrfs_fs_info * fs_info,int all)801 static void dump_devs(struct btrfs_fs_info *fs_info, int all)
802 {
803 struct btrfs_device *device;
804 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
805 unsigned long index;
806 int ret;
807 int i;
808 int j;
809 int cnt;
810
811 spin_lock(&fs_info->reada_lock);
812 list_for_each_entry(device, &fs_devices->devices, dev_list) {
813 btrfs_debug(fs_info, "dev %lld has %d in flight", device->devid,
814 atomic_read(&device->reada_in_flight));
815 index = 0;
816 while (1) {
817 struct reada_zone *zone;
818 ret = radix_tree_gang_lookup(&device->reada_zones,
819 (void **)&zone, index, 1);
820 if (ret == 0)
821 break;
822 pr_debug(" zone %llu-%llu elems %llu locked %d devs",
823 zone->start, zone->end, zone->elems,
824 zone->locked);
825 for (j = 0; j < zone->ndevs; ++j) {
826 pr_cont(" %lld",
827 zone->devs[j]->devid);
828 }
829 if (device->reada_curr_zone == zone)
830 pr_cont(" curr off %llu",
831 device->reada_next - zone->start);
832 pr_cont("\n");
833 index = (zone->end >> PAGE_SHIFT) + 1;
834 }
835 cnt = 0;
836 index = 0;
837 while (all) {
838 struct reada_extent *re = NULL;
839
840 ret = radix_tree_gang_lookup(&device->reada_extents,
841 (void **)&re, index, 1);
842 if (ret == 0)
843 break;
844 pr_debug(" re: logical %llu size %u empty %d scheduled %d",
845 re->logical, fs_info->nodesize,
846 list_empty(&re->extctl), re->scheduled);
847
848 for (i = 0; i < re->nzones; ++i) {
849 pr_cont(" zone %llu-%llu devs",
850 re->zones[i]->start,
851 re->zones[i]->end);
852 for (j = 0; j < re->zones[i]->ndevs; ++j) {
853 pr_cont(" %lld",
854 re->zones[i]->devs[j]->devid);
855 }
856 }
857 pr_cont("\n");
858 index = (re->logical >> PAGE_SHIFT) + 1;
859 if (++cnt > 15)
860 break;
861 }
862 }
863
864 index = 0;
865 cnt = 0;
866 while (all) {
867 struct reada_extent *re = NULL;
868
869 ret = radix_tree_gang_lookup(&fs_info->reada_tree, (void **)&re,
870 index, 1);
871 if (ret == 0)
872 break;
873 if (!re->scheduled) {
874 index = (re->logical >> PAGE_SHIFT) + 1;
875 continue;
876 }
877 pr_debug("re: logical %llu size %u list empty %d scheduled %d",
878 re->logical, fs_info->nodesize,
879 list_empty(&re->extctl), re->scheduled);
880 for (i = 0; i < re->nzones; ++i) {
881 pr_cont(" zone %llu-%llu devs",
882 re->zones[i]->start,
883 re->zones[i]->end);
884 for (j = 0; j < re->zones[i]->ndevs; ++j) {
885 pr_cont(" %lld",
886 re->zones[i]->devs[j]->devid);
887 }
888 }
889 pr_cont("\n");
890 index = (re->logical >> PAGE_SHIFT) + 1;
891 }
892 spin_unlock(&fs_info->reada_lock);
893 }
894 #endif
895
896 /*
897 * interface
898 */
btrfs_reada_add(struct btrfs_root * root,struct btrfs_key * key_start,struct btrfs_key * key_end)899 struct reada_control *btrfs_reada_add(struct btrfs_root *root,
900 struct btrfs_key *key_start, struct btrfs_key *key_end)
901 {
902 struct reada_control *rc;
903 u64 start;
904 u64 generation;
905 int ret;
906 struct extent_buffer *node;
907 static struct btrfs_key max_key = {
908 .objectid = (u64)-1,
909 .type = (u8)-1,
910 .offset = (u64)-1
911 };
912
913 rc = kzalloc(sizeof(*rc), GFP_KERNEL);
914 if (!rc)
915 return ERR_PTR(-ENOMEM);
916
917 rc->fs_info = root->fs_info;
918 rc->key_start = *key_start;
919 rc->key_end = *key_end;
920 atomic_set(&rc->elems, 0);
921 init_waitqueue_head(&rc->wait);
922 kref_init(&rc->refcnt);
923 kref_get(&rc->refcnt); /* one ref for having elements */
924
925 node = btrfs_root_node(root);
926 start = node->start;
927 generation = btrfs_header_generation(node);
928 free_extent_buffer(node);
929
930 ret = reada_add_block(rc, start, &max_key, generation);
931 if (ret) {
932 kfree(rc);
933 return ERR_PTR(ret);
934 }
935
936 reada_start_machine(root->fs_info);
937
938 return rc;
939 }
940
941 #ifdef DEBUG
btrfs_reada_wait(void * handle)942 int btrfs_reada_wait(void *handle)
943 {
944 struct reada_control *rc = handle;
945 struct btrfs_fs_info *fs_info = rc->fs_info;
946
947 while (atomic_read(&rc->elems)) {
948 if (!atomic_read(&fs_info->reada_works_cnt))
949 reada_start_machine(fs_info);
950 wait_event_timeout(rc->wait, atomic_read(&rc->elems) == 0,
951 5 * HZ);
952 dump_devs(fs_info, atomic_read(&rc->elems) < 10 ? 1 : 0);
953 }
954
955 dump_devs(fs_info, atomic_read(&rc->elems) < 10 ? 1 : 0);
956
957 kref_put(&rc->refcnt, reada_control_release);
958
959 return 0;
960 }
961 #else
btrfs_reada_wait(void * handle)962 int btrfs_reada_wait(void *handle)
963 {
964 struct reada_control *rc = handle;
965 struct btrfs_fs_info *fs_info = rc->fs_info;
966
967 while (atomic_read(&rc->elems)) {
968 if (!atomic_read(&fs_info->reada_works_cnt))
969 reada_start_machine(fs_info);
970 wait_event_timeout(rc->wait, atomic_read(&rc->elems) == 0,
971 (HZ + 9) / 10);
972 }
973
974 kref_put(&rc->refcnt, reada_control_release);
975
976 return 0;
977 }
978 #endif
979
btrfs_reada_detach(void * handle)980 void btrfs_reada_detach(void *handle)
981 {
982 struct reada_control *rc = handle;
983
984 kref_put(&rc->refcnt, reada_control_release);
985 }
986